UC San Diego

The need to acquire information about the variability in the world is paramount to optimal behavior yet it is not understood how this occurs on long timescales. Using the nematode Caenorhabditis elegans, I developed a novel learning paradigm in which the animals utilize their previous experience to guide their exploration for new sources of food. Using a dimensionality reduction technique, I find that the animals are searching based on the variability in observed food, and the time course over which they are learning. I also investigate the neural circuitry that underlies this behavior and the mechanisms by which plasticity occurs - via dopamine and CREB. Dopamine acts on two distinct D1-like dopamine receptors, one on a sensory neuron and the other on its postsynaptic interneuron where CREB is also acting. The amount of CREB in the cell controls the rate at which the animal learns about the environment. Further, the sensory neurons which detect this variability are specialized for the task and only respond to large fluctuations in observed bacteria. It is additionally unclear how to optimally use the available information. I utilize model of optimal information-seeking behavior to show that the optimal behavior seeks reward in the local area for a finite amount of time before moving to a new area when the reliability of information about the current environment becomes low. Despite the model using a local decision rule, there is an emergent global change in behavior from local to global search. This optimal behavior can be approximated by a drift-diffusion model of decision-making, suggesting a deep connection between previous models of optimality and psychological theories of decision-making. Crucially, the behavior predicted by the model matches the behavior observed in C. elegans